Build Configurations
NuttX can be built in three different configurations: (1) as a FLAT build where all of the code resides in a common address space, (2) as a PROTECTED build where a memory protection unit (MPU) is used to separate the memory into priveledged memory for the OS and unpriveledged memory for all applications, or (3) as a KERNEL build where a memory management unit (MMU) is used place the OS in a priveledged address space and to place to task (or process) in its own virtual address space.
In the last two configurations, applications reside outside of the OS address space and in all configurations applictions do not have have access to any internal resources of the OS.
More information about these build configurations can be found on the Memory Configuration Wiki page.
Thread Types
NuttX supports three classes of threads: tasks, pthreads, and kernel threads. tasks and pthreads are both application threads and are distinguished by some usage semantics and by their hierarchical relationship. tasks are created via several different mechanisms: task_create(), task_spawn(), execv(), posix_spawn(), and others. Tasks may then create pthreads using pthread_create().
More information about tasks and pthreads can be found on the NuttX Tasking Wiki page.
Kernel Threads
Kernel threads are really like tasks except that they run inside the operating system and are started with kernel_thread() which is prototyped in include/nuttx/kthread.h. The differ from tasks in that (1) in PROTECTED and KERNEL builds, they have full supervisor priveledges, and (2) they have full access to all internal OS resources.
In order to build the task into the OS as a kernel thread, you simply have to: (1) place the kernel thread code in your board source code directory, and (2) start it with kernel_thread() in your board bring-up logic. There a few examples of this in the NuttX source tree. Here is one: https://bitbucket.org/nuttx/nuttx/src/master/boards/arm/stm32/viewtool-stm32f107/src/stm32_highpri.c
So that is another trick that you can use to architecture optimal solutions: Create parts of your applications as kernel threads: They need to reside in your board/src directory and the need to be started with kernel_thread() in your board bring-up logic. And that is it.