A reduction of the algorithmic complexity of the fuzzy inference engine has the following property: the inputs (the fuzzy rules and the fuzzy facts) can be divided in two parts, one being relatively constant for a long a time (the fuzzy rule or the knowledge model) when it is compared to the second part (the fuzzy facts) for every inference cycle. The occurrence of certain transformations over the constant part makes sense, in order to decrease the solution procurement time, in the case that the second part varies, but it is known at certain moments in time. The transformations attained in advance are called pre-processing or knowledge compilation. The use of variables in a Business Rule Management System knowledge representation allows factorising knowledge, like in classical knowledge based systems. The language of the first-degree predicates facilitates the formulation of complex knowledge in a rigorous way, imposing appropriate reasoning techniques. It is, thus, necessary to define the description method of fuzzy knowledge, to justify the knowledge exploiting efficiency when the compiling technique is used, to present the inference engine and highlight the functional features of the pattern matching and the state space processes. This paper presents the main results of our project PR356 for designing a compiler for fuzzy knowledge, like Rete compiler, that comprises two main components: a static fuzzy discrimination structure (Fuzzy Unification Tree) and the Fuzzy Variables Linking Network. There are also presented the features of the elementary pattern matching process that is based on the compiled structure of fuzzy knowledge. We developed fuzzy discrimination algorithms for Distributed Knowledge Management Systems (DKMSs). The implementations have been elaborated in a prototype system FRCOM (Fuzzy Rule COMpiler).