Feature selection in pattern recognition is a problem whose space complexity grows exponentially regarding the number of attributes in a dataset. There are several hardware implementations of algorithms for overcoming this complexity. These hardware architectures relay on a software component for filtering irreducible features subsets, which is a computationally complex task. In this paper, a new hardware module for the filtering process is presented. The main advantage of this new architecture is that no additional time is required for hardware execution whilst the software component is no longer needed. Experimental results show that the runtime magnitude order for software is the same as for hardware in some cases. The proposed architecture is algorithm independent and may lead to smaller hardware realizations than previous architectures.