Many applications require the evaluation of some function through polynomial approximation. This article details an architecture generator for this class of problems that improves upon the literature in two aspects. Firstly, it benefits from recent advances related to constrained-coefficient polynomial approximation. Secondly, it refines the error analysis of polynomial evaluation to reduce the size of the multipliers used. As a result, architectures for evaluating arbitrary functions with precisions up to 64 bits, making efficient use of the resources of recent FPGAs, can be obtained in seconds. An open-source implementation is provided in the FloPoCo project.