This paper presents a new approach to the automatic identification of objects / classes from a system specification. The methodology is aimed at the development of Distributed Real-Time Systems (DRTS), specially those conceived for industrial automation applications. UML is used as the modeling language in conjunction with an extended version of traditional DFD (E-DFD), that conveys information on flow type (data or control) and on process complexity and timing constraints. E-DFDs are mapped into a graph and a tool is used to carry out an automatic graph partition, which allows the identification of a set of objects that constitute, from the design and test points of view, the 'best-fitted' architecture. To validate the proposed methodology, a case study is presented and analyzed, comparing two design strategies.