Linux kernels are made up of many modules, but is written in source codes(english), and has to be compiled(binary). When you do the compilation, you have to select CPU platform first. Which is now more than i386(16 bit/32 bit) but 32 bit(i486/i586), and 64 bit(i686 and others) Then you select drivers to be included. Last, you select networks to be included

You can learn very fast by studying Gentoo and T2 systems; and how they do it. When you selected the functional feature, you will be selecting sound and video drivers and other computer component drivers as well. Today, Xserver obsoleted all the other run levels, but Linux without automation still need startx.

Typically, i386 kernel 2.4.34 compiled is about 37 mb, and adding kernel 2.6.22 will have 46 mb. Many parts of Linux kernels are not usable in your own computer. But the most used Linux codes are only 4.5 mb, so the application execution memory size can be 9/16 mb in dram(data memory requirement is 2/3 times the execution instruction codes), ample for prefetch cached data operations. Typical microkernel operating system uses only 178,000 bytes(direct commands) in place of 4.5 mb of Linux monolithical logic(if then else) driven operations.

If you use some one else's base(distro), then you are stuck with their Linux kernel compilations, and you just add desktop and applications(package system such as RPM, etc.) in you installation script in your bootloader, which is lilo or grub.

So, many of us use cheat codes in bootloaders to speed up installation script by preselection of features before booting. this eventually will lead to speed up boot process tremendously; if distros use config data immediately without going thru installation script. Config file data can be preselected from the 4 digit codes of computer components in the bios installation data?