Until recently, Cloud Computing has not been viewed as a viable computational platform for HPC applications. This talk reviews the characteristics of both HPC and cloud architectures in the context of design and implementation of HPC as a Service. A novel methodology is presented for building and provisioning a secure high performance cloud layer for HPC applications and workflows. This layer can span X86, Power and other hardware, as well as different software environments. This paradigm may provide an option for better matching of a computing architecture with HPC applications and workflows. It also allows for safe experimentation with new schedulers and operating systems in a production HPC environment. A proof of concept bare-metal design is implemented using a Blue Gene/P supercomputer at Argonne's Advanced Leadership Computing Facility. Also discussed is future work including possible extension of this paradigm to Blue Gene/Q.