Coresets are small, weighted datasets that can be used as proxies of the original dataset in performing machine learning tasks. While previous coreset construction algorithms are tailor-made for specific machine learning tasks, we proved that k-clustering based coreset provides guaranteed performance for all machine learning tasks with sufficiently continuous cost functions. This is the first paper for Hanlin and the first deliverable for our coreset series of work. Congratulations and look forward to more and better products in coming years!