Generation of models from existing models composition: An application to agrarian sciences