While cloud computing has been broadly adopted, companies that deal with sensitive data are still reluctant to do so due to privacy concerns or legal restrictions. Vulnerabilities in complex cloud infrastructures, resource sharing among tenants, and malicious insiders pose a real threat to the confidentiality and integrity of sensitive customer data. In recent years trusted execution environments (TEEs), hardware-enforced isolated regions that can protect code and data from the rest of the system, have become available as part of commodity CPUs. However, designing applications for the execution within TEEs requires careful consideration of the elevated threats that come with running in a fully untrusted environment. Interaction with the environment should be minimised, but some cooperation with the untrusted host is required, e.g. for disk and network I/O, via a host interface. Implementing this interface while maintaining the security of sensitive application code and data is a fundamental challenge. This thesis addresses this challenge and discusses how TEEs can be leveraged to secure existing applications efficiently and effectively in untrusted environments. We explore this in the context of three systems that deal with the protection of TEE applications and their host interfaces: SGX-LKL is a library operating system that can run full unmodified applications within TEEs with a minimal general-purpose host interface. By providing broad system support inside the TEE, the reliance on the untrusted host can be reduced to a minimal set of low-level operations that cannot be performed inside the enclave. SGX-LKL provides transparent protection of the host interface and for both disk and network I/O. Glamdring is a framework for the semi-automated partitioning of TEE applications into an untrusted and a trusted compartment. Based on source-level annotations, it uses either dynamic or static code analysis to identify sensitive parts of an application. Taking into account the objectives of a small TCB size and low host interface complexity, it defines an application-specific host interface and generates partitioned application code. EnclaveDB is a secure database using Intel SGX based on a partitioned in-memory database engine. The core of EnclaveDB is its logging and recovery protocol for transaction durability. For this, it relies on the database log managed and persisted by the untrusted database server. EnclaveDB protects against advanced host interface attacks and ensures the confidentiality, integrity, and freshness of sensitive data.