In this paper, a methodology for the decomposition of dynamical systems is presented. The methodology uses two decomposition strategies seeking to reduce the multiple solutions and heuristics that input-output pairing methods usually yield. The first strategy quantifies the interaction between the inputs and outputs of the system. Therefore from a large set of inputs and outputs it will be selected only the input-output pairings with a strong dependence and the remaining are discarded. The second strategy allows to obtain subsystems with local relevant states and make sure that each of them contains the adequate input-output pairings. Two typical benchmarks are used to test the methodology application, including the physical construction of a thermal benchmark. For each benchmark, distributed state estimation was performed. Results show that the use of the proposed methodology can help in the decomposition of a real large scale dynamical system and to improve the state estimation in decentralized and distributed schemes.