I would start off with Danny Hillis' book "The Pattern On The Stone", which is very well written, and has a great introduction to how to make working computer logic from almost anything, included TinkerToys (Danny and Brian Silverman made a whole TIC TAC TOE playing computer from TinkerToys years ago (check it out online).
The progression has some parallels to going from 6 simple chemical elements to eventually define the architecture of a living -- and maybe thinking -- live creature. Computers are simpler by far, but still have a fair amount of detail, and a couple of sticking places (for example the design of the control elements).
I think -- after some physical manipulations -- I'd then get the kid to make some simulated logic gates in a graphical language like Scratch or GP or Etoys. These will allow a few simple things to be made (for example it is easier to make a flipflop and see why it works this way, and to see how one would make it from physical materials).
I think I'd then try to find the simplest logic CAD system and simulator on a personal computer and do some simple organizations that are completely understandable: go from flip flips to making an array that can act as a memory, add address lines and bus lines, etc.
Another fairly simple design and build project is to make a simple ALU that can do a few things and hook it to the address lines.
So far we have something that is about the complexity of an old-time built from scratch model airplane: in the range of some 12 year olds and most 14 year olds, and a very few special 10 year olds.
A working memory and an ALU with busses can be used to show how a typical computer gets works from memory, does something with them and puts them back into memory. This is already quite a project. Note that every part of it is very understandable -- what's tough is the amount of stuff that has to be assembled to make this.
The next level is to build the control part, and this is the first organizational part that is a bit more tricky, and a bit more complex to design and understand. It might be more worthwhile to look at the simplest pedagogical computers rather than trying to build one here.
There are courses -- mostly for college -- that do this entire chain of being with perhaps high school students onwards. A really good one is by Nisan and Schocken at Hebrew University of Jerusalem. It includes a CAD/SIM system and you can do everything from scratch.