In the field of machine learning, one of the famous algorithms is support vector machines (SVM). It has been used to solve classification and as well as regression problems. The most crucial part of SVM is the kernels. There are several widely used kernel functions. A well-designed kernel will help in increasing the accuracy of classification by SVM. Main aim of this paper is analysing and designing of customized kernels. After exploration of all the kernels, a new kernel function has been designed. The proposed work includes the theoretical analysis of the interpretation of kernels to feature space. Further, an experiment was done on four different data sets collected from UCI. The observed improvements identified with the proposed kernel are compared with the existing linear, RBF and polynomial kernels. However, these best results are not universal and not fixed.