For example to dock a silver molecule to a receptor in various size ranges from 10nm,20nm,50nm,100nm e.t.c. Is it possible to do so in the docking software in place of ligands?
Although possible in principle, at least three issues arise. The first one is with the algorithm for poses (or relative orientation) prediction, designed for much smaller "ligands". The second one comes from the force fields not necessarily updated for metal atoms forming the nanoparticle and the third one deals with emergent properties of nanoparticles. These sizes of particles confer special chemical characteristics, not necessarily described by any force field. I would recommend performing some previous docking experiments with at least a couple of already known structures in order to test the performance of AutoDock with these potential problems.
You could use a semi-empirical or ab initio calculation.
The MOPAC software (semi-empirical) has included a keyword to work with large systems like proteins, or you can use the Quantum Mechanics/Molecular Mechanic (QM/MM) mix. Also the Fragment Molecular Orbital (FMO) is capable to do calculations for large systems using DFT, HF, etc. (it uses a "divide and conquer" technique).
Any of them will work much, much better than docking (even for ligand/protein).