The Langlands functoriality conjecture implies that automorphic $L$-functions belong to the Selberg class, but not the other way (i.e. the other direction is not known to follow from this conjecture). Regarding to Rankin-Selberg convolutions, I don't think that this operation has been defined precisely for the Selberg class. There is a subtlety at the ramified primes: one usually refers to the algebraic classification of the underlying local representations, which is itself part of the Langlands program.