Software cost estimate is important not only to project plans but also to software budgeting. Although more and more modules and methods are presented, it seems that Artificial Intelligence is the only way to form relatively precise solutions. In this paper, we focus on Nesma, a kind of popular Function Point Measurement, and present 5 main paradigms to define heuristic rules to split the software into two layers, which are Pricing Objects and Measuring Objects, for precise mapping for ILF, EIF, EI, EQ, EO. Moreover, after large-scale information projects are set as training sets, we use paradigms to improve the performance of the original LSTM-CRF. It is illustrated that paradigms-based LSTM-CRF has achieved significant results in controlling the increase in accuracy and precision rate while improving performance. However, the recall rate remains high. Compared with experts' manual auditing results, the quantity and quality of the training samples and manually labeled texts still need to be improved.