Finite automata are used to encode geometric figures, functions and can be used for image compression and processing. The original approach is to represent each point of a figure (a graph of a function) in $\mathbb{R}^n$ as a convolution of its $n$ coordinates written in some base. Then a figure is said to be encoded as a finite automaton if the set of convolutions corresponding to points in this figure is accepted by a finite automaton. Jurgensen, Staiger and Yamasaki showed that the only continuously differentiable functions which can be encoded as a finite automaton in this way are linear. In this paper we propose a representation which enables to encode piecewise polynomial functions with arbitrary degrees of smoothness that substantially extends a family of functions which can be encoded as finite automata. This representation naturally comes from the framework of hierarchical tensor product B-splines utilized in numerical computational geometry. We show that finite automata provide a simple tool suitable for solving computational problem arising in this framework including the case when the support of a function is unbounded.