This paper addresses the problem of using Support Vector Machines (SVMs), a type of intelligent system for classification of non-linear data sets. The idea is to use the Kernel function and utilize its full potential to enhance the efficiency of classification a linear SVMs can do for non-linear dataset. By dividing the feature space of the data into sub-categories of linearly separable data-points and executing a linear SVM for each of these categories, the one-versus-one and one-versus-rest techniques are applied here. This is achieved by understanding the overlaps of the data and then segregating it category wise. The variable of kernel trick is iterated to achieve a higher level of accuracy. The data used for this study is of learning disability where the purpose is to identify the type of learning disability from the available 7 combinations. The experimental findings show that the proposed approach is equally effective at making predictions as a linear SVM.