Yes, the exponent can be any odd value greater than 1.
We usually insist on prime values of $e$; why is this? Well, it does make generating key pairs a bit easier.
When we generate an RSA key pair, standard practice is to pick a public exponent $e$, and then select primes $p$ and $q$ that work with that $e$. And, by work, I mean that $gcd(e, p-1) = 1$ and $gcd(e, q-1) = 1$. Now, if $e$ is prime, these requirements are equivalent to $p \ne 1 \pmod e$ and $q \ne 1 \pmod e$; these are easier to test than explicitly computing the gcd (and if you are using sieving to find your primes, it's easy to stir in this additional relation into the sieve). If $e$ wasn't prime, we'd either need to explicitly compute the gcd, or use the relations $p \ne 1 \bmod f$ for all prime factors $f$ of $e$. Neither would be infeasible (or even all that difficult); however the prime case is easier.
Now, there is a counterargument in favor of composite values of $e$; it's that the RSA problem with a composite $e$ can be reduced to the RSA problem with all the prime factors of $e$; for example, the RSA problem with $e=33$ is hard if either the RSA problem with $e=3$ or the RSA problem with $e=11$ is hard. I've never heard anyone make this argument (possibly because we have no evidence that any odd $e > 1$ makes the RSA problem easier than any other)