In this paper we propose a novel procedure for obtaining a low order non-linear model of a large scale multi-phase, non-linear, reactive fluid flow systems. Our approach is based on the combinations of the methods of Proper Orthogonal Decomposition(POD), and non-linear System Identification (SID) techniques. The problem of non-linear model reduction is formulated as parameter estimation problem. In the first step POD is used to separate the spatial and temporal patterns and in the second step a model structure and it's parameters of linear and of non-linear polynomial type are identified to approximate the temporal patterns obtained by the POD in the first step. The proposed model structure treats POD modal coefficients as states rather than outputs of the identified model. The state space matrices which happens to be the parameters of a black-box to be identified, comes linearly in parameter estimation process. For the same reason, Ordinary Least Square (OLS) method is used to estimate the model parameters. The simplicity and reliability of the proposed method gives computationally very efficient linear and non-linear low order models for extremely large scale processes. The method is of generic nature. The efficiency of proposed approach is illustrated on a very large scale benchmark problem depicting Industrial Glass Manufacturing Process (IGMP). The results show good performance of the proposed method.