Discriminative and generative machine learning algorithms have been successfully used in different classification tasks during the last several decades. They both have some advantages and disadvantages and depending on a problem, one type of algorithm performs better than the other one. In this paper we contribute to the research of combination of both approaches and propose literature based a hybrid discriminative-generative generic model. Also, we propose hybrid model structure finding and building a new algorithm. We present theoretical and practical advantages of the hybrid model over its consisting algorithms, efficiency of the model structure finding algorithm, then perform experiments and compare results.