In recent years, deep learning has achieved noteworthy success in various applications. The existing DL libraries widely used by researchers are too complex and challenging to modify the code functions with the custom mathematical formula for research purposes. Thus, we propose a software library related to deep learning that facilitates researchers and practitioners in academia and industry to mutate the training model function with the specific equation. In this paper, we address preparation before developing a software library for deep learning to inspire the researcher who wants to develop a software library. They can utilize our research methodology step by step to conduct the software library architecture design process. We perform four main phases with different activities that execute sequentially to construct a software library intended for deep learning implementation into code practically, and we produce 11 documents to facilitate the transfer of detailed knowledge from us as the pioneer developer of a software library for deep learning that will be part of Analytical Library Intelligent-computing (ALI) to the following developers.