I recently created my own planetarium/ephemeris, here are some (rambling) thoughts:
There are two kinds of time that can be used as input to your program: Ephemeris time (coordinate time), and TAI/UTC (proper atomic time). Coordinate time is the independent variable of the equations of motion.
The input to the mentioned VSOP and JPL ephemerides is a coordinate time, which is not the same as the UTC date and time on Earth. If you wish calculate the positions of the planets as a function of the actual time, you must first convert the time into a suitable coordinate time. TE405 is an open source project that converts TAI to the Ephemeris time used in the JPL DE405 ephemeris.
If you want reasonably accurate results, you must numerically integrate the Einstein-Infeld-Hoffmann equations of motion, including the harmonic coefficients of the gravitational potentials of the Sun, Mercury, Earth, Jupiter, and Saturn. You will need to integrate the orbits of the larger minor planets (Ceres, Pallas, Vesta) as well. Numerical integration of the relativistic equations is very computationally intensive, and requires quad precision. NASA uses special hardware, everyone else uses a high precision software library. Unless you know GR, it's best just to use a Chebyshev approximation
Precision is real problem. Intel processors have 80 bit FP registers, but Windows only supports 64 bit FP types. By coding in assembler you can retain that precision.
Once you have the positions of the planets it's not that much more work to add a 3d visualization of the solar system using DX or OPENGL. For a realistic background use the Hipparcos, or Yale bright star catalogs. After that, go to the minor planet database and download the orbits of the first 300,000-400,000 minor planets in CSV format, any modern GPU will be able to render (but not animate) them in real time. Get a 3d monitor and prepare to be blown away.
If you want to reproduce real astronomical observations you will need to account for precession/nutation/polar motion, etc.
To replicate Lunar ranging experiments, use the GTOPO30 digital elevation model.