Yes, of course it would deter computers...
... if it were done as substitution and every instantiation of a symbol had a new meaning. This would avoid frequency analysis.
But you are going to be drawing a lot of pictures, and your correspondent is going to need a thick book of hand-drawn symbols if you have a lot of traffic (will each of your hand-drawn symbols look exactly alike each time you draw it?).
Moreover, you now have a physical security problem. And keep in mind that this isn't art class--it's cryptography. But what you propose was widely done in the past, especially in the sixteenth century, and it was serious business.
However, the sixteenth century is over.
If you go ahead anyway, you should not reuse your symbols if they map one-to-one to an individual phoneme (if you only use substitution and not transposition). If you are ambitious and you enjoy drawing, you could create a system in which you map a phoneme to a different pictogram each time, which would surely be effective--but, again, this would all be time-consuming.
Nothing is gained by using phonemes, and perhaps nothing is gained by drawing the symbols by hand (unless you do not trust your computer). Just use ASCII or extended ASCII.
Let's skip the fact that not only will the language have to be agreed upon, but also the dialect. Doing this for English (at least) is going to become a rat's nest of uncertainty very quickly. Which system of phonemes to use also needs to be determined, and this will not be easy. Will you use IPA? Will you say there are 42 or 44 phonemes? Which English dictionary will you use to determine/check the pronunciation? What is the pronunciation? If you look at various English dictionaries, you might be surprised to discover that they sometimes disagree about proper pronunciation.
Anyway, let's consider this ciphertext (plaintext "go up") based on English phonemes:
β ┤ Ω ♯
Is a computer going to be able to break it without knowledge of your code book? No. Without knowledge of your preshared secrets, cryptanalysis is going to hit a wall.
Is this cryptography? Yes. According to Claude Shannon:
... “true” secrecy systems where the meaning of the message is
concealed by cipher, code, etc., although its existence is not hidden,
and the enemy is assumed to have any special equipment necessary to
intercept and record the transmitted signal.
But do we want to go back to this the kind of primitive cryptography? Other methods exist which are much easier and far better--and may offer an integrity check and authentication.