Asymptotic description of stochastic neural networks. I. Existence of a large deviation principle