The increase in data traffic over the past years is predicted to continue more aggressively in the years to come. However, traditional methods such as increasing the amount of spectrum or deploying more base stations are no longer sufficient to cope with the traffic growth. Caching is hence recognized as one of the most effective means to improve the performance of Internet services by bringing content closer to the end-users. Although the benefits of in-network content caching has been demonstrated in various contexts, they introduce new challenges in terms of modeling and analyzing network performance. Building on analytical results for Time-To-Live caches and the flexibility they provide in modeling caches, this thesis investigates various aspects in which caching affects network design and performance. The complexity of making optimal routing and content placement decisions is studied first. Showing the infeasibility of implementing the optimal strategy, low-complexity techniques are developed to achieve near-optimal performance in terms of the delay observed by end-users. The problem of differentiated cache services is studied next with the question ``how can Content Distribution Networks implement caching policies to provide differentiated services to different content providers?''. A utility-maximization framework is formulated to design caching policies with fairness considerations and implications on market economy for cache service providers and content publishers. An online algorithm is also developed with the purpose of implementing the utility-based cache policies with no a priori information on the number of contents and file popularities. This thesis also analyzes caches in conjunction with data structures, e.g. Pending Interest Table, proposed in the future Internet architecture designs such as Named Data Networking. The analysis provides the means to understand system performance under different circumstances, and develop techniques to achieve optimal performance.