Parameter estimation is a common challenge that arises in the domain of computational scientific modeling. Agent-based models offer particular challenges in this regard, and many solutions are too computationally intense and scale with the number of parameters. In this paper, we propose knowledge-based function approximation methods to deal with this problem in agent-based modeling. Our method is implemented within the VERA modeling system, and we show the validity of our methods using an internal model as well as an external model.