You don't say what kind of elements you plan to use.
If they're 8 node linear bricks or 20 node quadratic bricks it's an "easy" problem: just create cubic regions for each different material and create your mesh.
I'm guessing that the interphase/interface layer is "thin". You'll have some aspect ratio/transtion problems to sort in the matrix and inclusion models.
This can also be done with a spherical inclusion with a shell interface/interphase between inclusion and matrix, but it'd be easier with tetrahedral elements and an automatic octree meshing solution.
You can exploit planes of symmetry: three in this case. Symmetry/antisymmetry boundary conditions can help you combine different loading methods.
You should not be using boolean operations to create this. Just create a cube with regions for each material. It's straightforward.
Can't show in 3D, but 2D will give you a hint:
+--------+---+---------+
| 1 | 2 | 3 |
| | | |
+--------+---+---------+
| 4 | 5 | 6 |
+--------+---+---------+
| 7 | 8 | 9 |
+--------+-------------+
- Inclusion: region 7
- Interface: regions 4, 5, and 8
- Matrix : regions 1, 2, 3, 6, and 9.
Set material properties in each region appropriately.
Add boundary conditions to the left vertical and bottom horizontal surfaces. Loadings are applied to right vertical and top horizontal surfaces.
Now you should see what I'm getting at: the interface regions are probably thin, and their material less stiff than the inclusion or matrix. You'll have very small elements in that area in order to maintain good aspect ratios, which means lots of matrix and inclusion elements.
I'd recommend doing a simpler case in 2D, like the one I have here, to get a sense for the behavior before you try 3D. You'll be able to turn it around quickly and even check convergence before you tackle the bigger problem.