The paper lays the foundations for the design of a programmable controller capable of handling nonlinear behavior of chemical reactors or other processes, where the model available for their dynamics could be not completely known. We elaborate on results about the approximation of nonlinear systems and their optimal control, and then combine the theoretical conclusions together with numerical strategies for identification, filtering, and updating, which take into account possible changes in the structure or the values of the coefficients of the original system.