Neural Networks for years have been used in various applied mathematical algorithms. Modern complex problems have big dimensionality and huge number of free variables. In most cases they require hardware acceleration or a cluster to make the computations fast enough. There are too many ways how the final, enterprise system can be built by combining existing neural network and parallel computation technologies. The problem of choice is getting more complicated due to the need to select hardware brand. "Neuromathematics" is another project aimed to facilitate rapid development of the software products devoted to solve large scale math problems via neural network algorithms and efficiently utilize specialized hardware or cluster. The power of interfaces (C++ abstract classes) was applied to separate the main computation algorithms from technological specifics, leaving flexibility in choosing the environment and keeping the applied task solution code C++ - efficient, clean and easy to debug. The interfaces of the system formed "the skeleton" of the platform, to which the different subsystems implementations could be plugged in and tried without the need of the applied problem code is even recompiled. In the context of several prior factors "Neuromathematics" approach was the most cost and time effective and extendable solution. The same approach may be useful to follow in the similar project contexts.