The mobile traffic grows every year. Resource requirements of mobile applications, as processing power and storage capacity, transformed the architecture of cellular networks into a centralized infrastructure, the C-RANs. On the other hand, providing resources close to base stations, as the multi-access edge computing suggests, allows an immediate processing for delay sensitive applications. In this work, we formulate as a mixed integer linear programming the retrieval of media contents through caches acting as virtual network functions. The optimization model minimizes both the media contents' retrieval cost and the number of instantiated virtual network functions. Results show that the characteristic that has most impact on caching is the high throughput between the node that stores the virtual network function images and the virtualized environment that hosts the image, where up to 70% of the requests are satisfied. When caches have low storage capacity and incurs high transferring delay cost for deployment, the most important characteristic is the throughput distribution between the virtualized environment and the base stations.