Conventional Non-Gaussianity metrics used for independent component analysis (ICA), may not sufficiently apply all needed information in efficient source reconstruction. This may happen due to limiting usage of only fourth moment of data. Other nonlinearities also, presuppose distribution for data leading to usage of certain moments in solving ICA problem. This issue of ICA has been addressed many times in literature for EEG and speech signals. In this paper, a novel algorithm has been proposed to handle cases with unknown probability distributions led to the vagueness of nonlinearity as a priori. The approach is to first, design a more generalized objective function of Non-Gaussianity for projecting all data moments in comparable scale with respect to each other; then derive gradient of proposed objective function for updating mixing matrix of one component. And finally turn one-component case to a symmetric multi-component gradient function equipped with an orthogonality enforcement update term. The proposed method has been evaluated against well-known blind source separation algorithms. It has shown its ability to overcome them all in terms of commonly used independency metrics. Also proposed approach is conducted in classification process of EEG speech imagery dataset as preprocessing source separation phase and manifested the highest accuracy in test data over other well-known BSS methods.