We present the modeling of dynamical systems and finding of their complexity indicators by the use of concepts from computation and information theories, within the framework of J.P. Crutchfield's theory of  ε-machines. A short formal outline of the  ε-machines is given. In this approach, dynamical systems are analyzed directly from the time series that is received from a properly adjusted measuring instrument. The paper serves also as a theoretical foundation for the future presentation of the DSA program that implements the  ε-machines modeling up to the stochastic finite automata level.