Due to ever increasing size of datasets with an extremely high number of attributes, mining of datasets turns out to be a universal scenario in various fields such as, industry, social and scientific areas. In order to search with very large databases, clustering is an important tasks required in various applications. Moreover, clustering algorithms that provide better scaling to very big data are significant and useful. When the size of the data become too big, it is unfeasible to store it in a single computer memory and also it consume more time for entire kernel matrix computation. In order to alleviate this problem, we propose a multiple kernel and a swarm-based MapReduce framework for big data clustering to improve the clustering accuracy. To map the input data, we implement a new clustering algorithm using tangential kernel and spherical kernel, which is used for calculating the relevant cluster centroid. Then, particle swarm based optimization (PSO) is utilized in reducer function, where the fitness value is calculated using Davis-Bouldin (DB) index. The proposed multi kernel swarm based MapReduce framework is experimented with skin and localization data sets. It can be concluded that, the proposed MKSMRF achieves higher clustering accuracy of 81% and 85% for localization and skin dataset respectively, as compared with other clustering algorithms.