Complex systems are assemblies of interacting components, which themselves could be sub-systems. Complex behavior of these systems is the consequence of the behavior of its individual components or sub-systems and the interaction between them. Information about the complex behavior of these systems can be gained either from observation or by computational simulation. Developing a model for computational simulation of a complex system requires models of the behavior of its components and the interaction between them. It is important to note that the behavior of the components of complex systems in isolation often differs considerably from the embedded behavior of those components in interaction with others within the complex system. In this conceptual paper we propose a new method for developing models of the in-situ behavior of the components of the complex systems from the observation of the stimulus-response behavior of the complex system itself. The proposed method is the adaptation and application in complex system of a method called Autoprogressive Algorithm (Autop) or Self-learning Simulation (SelfSim) that was developed and successfully applied in many problems in computational mechanics and engineering. All the previous applications of Autop and SelfSim in engineering have been problems that differ considerably from, and do not exhibit the same properties as in complex systems. In this paper we are exploring the potential and promise of the application of the Autop and SelfSim in complex systems. The behavior of the components of the complex systems and the interaction between them are often highly nonlinear, adaptive and stochastic. Moreover, most complex systems are nested; the components are themselves complex systems and their behavior is also complex. The nested structure can be at multi-levels. These observations make it extremely difficult to develop purely mathematical models for behavior of the components of complex systems and the interactions between them. The alternative is to use informational approach; to acquire and store, in a reusable form, the information about the behavior and the interactions of the components of complex systems. In this paper we are proposing a Hybrid Mathematical-Informational Method (HMIM) to use the existing mathematical models while supplementing them with informational method.