On the Consistency and Large Deviations of the Method of Empirical Means in Stochastic Programming Problems