The first step in running managed code in a process is to get the CLR loaded and initialized using a CLR host. Typically, a host consists of both managed code and unmanaged code. The managed code which executes in the default domain is usually responsible for creating the application domains in which the user code exists. All CLR hosts must contain unmanaged code because execution must begin in unmanaged code. The .NET Frameworks provides a set of unmanaged APIs that the host can use to configure the CLR, load the CLR into a process, load the hosts managed code into the default domain, and transition from the unmanaged code to the managed code.
The second step in running managed code in a process is to create application domains in which the user code will execute. The creator of the application domain can specify criteria which control code isolation, security, and loading of assemblies.
The third step in running managed code in a process is to execute user code in one or more application domains created in the previous step. All code that is run in the CLR must be part of an assembly. There are three options for loading assemblies. First, precompiled assemblies can be loaded from disk. Second, precompiled assemblies can be loaded from an array of bytes. Third, assemblies can be built dynamically in an application domain using the BCL Reflection Emit APIs.
Note. For an application launched from the command-line, the shell host executes the steps described above on behalf of the user and hides the complexity from the user.