Search meta-heuristics, such as tabu search and chaotic search, have been successfully used for solving NP-hard problems. These successes have led us to to be motivated to build a modular, reusable and adaptive model of the partially generalized searching system in a C++ library called neurosearch. In this paper, we explain the key concepts of neurosearch and how to support a VLSI netlist partitioning program. The library contains novel data structure management and basic functions for a move-based search using moves generated by a neural network. We expect that this contribution could accelerate the design-time for any move-based NP-hard problem solver construction. Moreover, it opens the possibilities of relatively independent cooperation among problem solving developers, neural network designers and neuron designers. The modular structure of the library also allows separate tuning programs, both for the search parameters and for the neural network parameters.