The Onomatopoeic Nature of Language
When the first caveman tried to inform his comrades of the presence of a bird in the vicinity, he probably pointed; this likely failed, as some of his associates, thinking he was indicating a woolly mammoth, probably fled. He then, most likely, tried flapping his hands and this worked but, as his club fell from his grip, ravaging several of his toes and preventing him from further pursuing the tasty member of the local ornis, he shouted and considered using his mouth instead. So, on later occasions, he most likely chirruped. In the course of these few hundreds of thousands of years, the invention of grammar and several other factors, the nature of which would exceed both the scope of this work and your patience, have mutated language (though several words, such as 'twit' are still clearly recognisable as being of imitative origin): sound is no longer used in prose to convey a simple meaning, but it has become one among the foremost instruments of poetry in Western Europe, due to its former association with music.
Many centuries having passed, and cavemen being now mostly employed as football hooligans, even the art of conveying meanings by the sheer sound has evolved somewhat; however, its original, imitative form often surfaces. The Futurist movement, which thought mechanical noises were the sweetest melody, produced several pieces meant to whir and rumble; here is the beginning of F.T.Marinetti's ode to a racing car:

Two things are remarkable in this piece: first, Marinetti multiplies the Rs in some words to create the effect of a revving of engines, and second, he associates them with unvoiced consonants, which sound 'inhuman'; a similar example in classic poetry is this line of Racine's, also quoted by Lausberg:

in which the abundance of phonetic Ss, coupled with the Fs of the central 'sifflent' produces a distinct sensation of reptiles slithering around. As with most poetic techniques, English evidence is hard to find; however, still on the subject of snakes, here is something by D. H. Lawrence:

There are, of course, cases of explicit onomatopoea (things like 'and the dog went bark, bark'), even by such authors as Shakespeare and Eliot, not to mention Poe, but these are generally too silly to be quoted even by the Present Author.

Subtler Effects

As the sonnet above dramatically demonstrates, the choice of sounds in poetry is not limited in its scope to imitating natural noises. There are quasi-mystical theories that actually think of connecting groups of sounds to unrelated abstract concept on a purely psychological basis (e.g., the 'F+L' group would be associated to moving lights, something that a medic acquainted with flukes and flatulence would strongly disagree with), and it is with great pleasure that we dismiss these claims as preposterous. But it is a fact, and one proven by thousands of examples, that there is a correlation between phonetics and the psychology of communication: anyone could quite see that the Italian word 'bacio' (in which the first consonant is pronounced with the lips, and the second forces the speaker to disclose them more) is more effective than the English word 'kiss' (in which the first consonant has the speaker open his mouth and bare his teeth, while the second makes him look as if he were growling) in conveying an amorous thought. Following this concept, based mostly on how the air touches one's vocal organs when sounds are pronounced, poets have built, since the earliest ages, a sort of easel from which vowels and, more often, consonants are chosen according to the mood of their work. The simplified version of this are the popular terms of euphony and cacophony; what is often labelled as the latter is an attempt at describing unpleasant realities; Arnaut Daniel, in a piece in which he defends a knight refusing a sexual practice he finds repulsive, starts:

Here the masculine ECX rhyme creates the effect of someone sneering at the end of each line; the one-stanza-one-rime choice enforces this, and so does the fact that several line endings are spondaic, forcing the reader to stop as if disgusted before the last word (this is actually a rhythmic device as described below).
E. Pound, in his Canto XLV, already quoted several times, uses the archaic -TH endings and several other means in order to show how usury affects economy (all right, all right, I admit the subject is just as unpoetic as it might get) by making the verses flow in a hindered, irksome way:

Baudelaire, and other symbolist poets achieve the same effect through nasal sounds; the reader can find example of this in a multitude of works.
On the other hand, some schools of poetry, the first of which was perhaps that of the 'Dolce stil novo' (sweet new style) in 13^th century Tuscany, believed that only matters of love were worthy of being treated (which is a bit limited, but still a great deal more interesting than the economy of the Middle Ages, if you ask me), and that this should be done using sounds that rolled easily off the tongue and pleased the ear. It was the first systematic application of what came later to be termed euphony, and which some poets, most notably Shelley, applied in a rather dumb way to just about everything. Here is an example by the Dolce Stil Novo master Cino da Pistoia (the form, an Italian ballad, is rare and obscure, and has not been described in this manual):

Cino tries to create a sensation of beatitude mostly by multiplying the diphthongs (in bold), and by avoiding the harsh sonority created by double consonants or by groups of them. Other typical rules of thumb used in this kind of writing is favouring certain sounds (typically, B, L, M, V) and shunning others (typically, TH, K, S); Cino doesn't really do that here, and uses the expression 'guardando', which, in corny Italian poetry is almost always replaced by 'mirando'.
As hinted above, some prudence should be used in moving from one language to the other; the sound 'R', for example, is only apparently the same in different languages: it actually is a very light, endearing sound in French (it's one of the actual reasons why most men find French girls sexy), a fairly neutral one in English and a forbiddingly harsh one in Italian. Furthermore, what we could term 'phonetic extremes', the sounds that are more difficult to pronounce, are generally limited to few languages: the TH of English does not exist in Italian, French or Provençal, and it is slightly different in Spanish; the -GL- sound of Italian is stronger than French or Spanish -ILL-, and has no equivalent whatsoever in English; twinned consonants are actually pronounced as such in Italian only, and so forth.
Things are even more complicated when it comes to vowels; the 'A' existing in most languages in the world, the 'velvety corset of springing flies' of Raimbaud, does not exist in English (some say it corresponds to the U sound in, say, 'cut', suggesting that they are either deaf of very stupid), the 'I' is different, double vowels are endemic to a few northern European languages, nasal ones exist in French but not in Provençal, in Portuguese but not in Spanish, in Icelandic but not in Swedish and so forth.
A good phonetics book (e.g., Lausberg's) can enlighten you more on this quite confusing point, which would, again, exceed the scope of this work; on the other hand, anyone familiar enough with his own language certainly does not need to be told which sounds are harsh and which are not. As for people that are not familiar with the language they read, they should be aware that they are going to miss more than they think.
It is therefore more interesting to consider a further example, and a British one at that, to see how the 'right' sounds are exploited. It is actually very difficult to eliminate entirely a series of words in favour of others; it is more reasonable to highlight them through their involvement in the homoeoprophoron or their position in the verse.

This first stanza of Blake's introduction to Songs of innocence is based much on L's (and on the fairly neutral P's) and contains a relatively huge number of diphthongs; the effect is that of a piece that could be read even by a child who isn't able to say all words properly. Note how this is achieved both by homoeoprophoron and lipograms (neither the short U nor the short I appear in this stanza).

Rhythm and Effect
If you care to reread the related chapter, you'll notice two things about the classic verse:

1. that it grants the writer a good deal of freedom
2. that where it does not, it imposes obvious, mechanical conditions

The structure of this poem is rather complex, but what the ear immediately perceives is that eleven of the twelve lines end with either of the two patterns UUSUSU or SUUUSU (which, on top of that, resemble each other). The only exception cleaves the piece in two just under the first occurrence of the words 'cuerpo' and 'alma', that appear again just below the ending. The global result oddly recalls a rondeau, with its lines divided in two groups of rhymes (replaced by the end-patterns) and its repeated hemistich (replaced by the double iteratio).
There is nothing new in tthe technique except for the fact it is actually applied to a good poem; for centuries, indeed, poetasters of all sorts have practised what is sometimes called barbaric meter, that is, the application of Greek stanzas to modern languages. Since typical Grecian verses have fixed stress patterns, at least in some lines, the effect of this is close to the piece above. Perhaps the most extensive attempt at this outside English, Latin and Greek, is the collection called 'Odi barbare', written in Italy by a pompous, patriotic fool called Carducci (who, as all pompous, patriotic fools, wallows in undeserved fame); here is the introduction (the rest is even sillier):

English poetry has always been proud of being barbaric, so this approach has been adopted from the beginning; iambic pentameter, in its dullest, most literal form, is a sequence of equal USUSUSUS patterns, as are the other examples of versification based on feet in chapter 1, the problem with it being that strictly single-pattern lines are just as trite, repetitive and, in the end, as uninteresting as their single-rhyme counterparts.
As with sounds, it is legitimate to wonder whether a particular rhythmic pattern can be associated with the imitation of natural phenomena or to a particular psychological state; this is a subtle question, and the answer is probably no. However, rhythms can be combined with accurate phonetic choices and used to strengthen their meaning. An easy parallel is the one with music: the length and volume of notes are fundamental, but it is very hard to write a minuet for drums.
We have seen before, especially with the homoeoprophoron, how this can be done; it is worth mentioning that the the most common exercise of all, especially among overly pelagic authors, is trying to use language to imitate the ebb and tide of the sea; among the many that did it there are Neruda, a pseudo-Raimbaut de Vaqueiras and Derek Walcott whose 'A sea-chantey' is concluded by these lines:

It is especially the last repeated line that is the most meaningful: the short stressed vowel of 'amen' suggests a wave breaking, the two long ones of 'calm' and 'waters' suggest the same retreating. As in ancient poetry, more than the actual arrangement of stressed and unstressed syllables, it is their length that is of paramount importance; this is a feature of the English language that should be considered more, especially by those that rave about Endymions, Adonaises and Chapman's Homers: but the dead cannot emend their stupidity.
Wonders can be done, however, even with Romance languages, in which all vowels are short: Antonio Machado manages to somewhat recreate the exasperating, intermittent buzzing of flies in his 'Las Moscas':

Pie Quebrado
A last thing worth mentioning before calling this a day is how, especially in a rhymed piece, the alternation of different verse lengths affects the general idea of a poem. It is one of the oldest habits of classical poetry to interrupt the regularity of successions of long verses with shorter ones. In Spain, already in the 13^th century there were pieces in which octasyllabic lines were interrupted by tetrasyllabic ones; these are called 'pies quebrados' (literally, 'broken feet'). Their application is still alive today; a pie quebrado sounds exclamatory, if not snappish: here is the next stanza of 'Las Moscas':

Other languages have 'estabilished' ways of doing the same thing. In Italian the endecasillabi are customarily mixed with lines of six metric syllables, French has the Rondeau form and even in English Donne fittingly uses pie quebrado, among others, in 'Womans constancy':

This example is particularly ingenious because of the irregular pattern that matches the fickleness of the poet's mistress; in particular the four-syllables line, of Spanish origin comes to highlight a sudden change of mind.
Note also how a long verse after a sequence of short ones has the effect of a comment, especially if it concludes a piece or stanza. As such it is used not only by Donne, but also in all of Spenser's Faerie Queene.

Now, Kind Reader, you who have followed me for so long through my toil, I release you: go, and write or, if such is not your call, go and read, teach or sleep metrically sound dreams. Conclusions are not for the enlightened to read: they are but a paltry thing, meant for perfectionists and editors, and those that just can't get enough of our idiosyncratic writing. Are you one? Why, then Ile fit you, too.