The questions I'd be asking myself are:
1. WHY am I writing the model ? what simulations am I going to use it in? what measurements do I need to make?
2. Which of the converter specifications are important in those simulations for those measurements?
3. What non-idealities am I depending on/ ?

Next I'd draw out a model schematic
Input characteristices - Transfer functions -(and its independent variables) and output characteristics.. a good reference for this is a charge-pump application note..  (ie from ADI, National, TI, Linear tech etc.. )

Then your about ready to start your implementation.. Which indep variables are block parameters and which are inputs (or simulator variables?)
Fclk, and Cout might be block parameters, but Vdd, and Vref might be input pins from external sources..
but of course its always good to start with the simplest model..

Assuming 3 stages of voltage doubling, why can't you just use Vout = pow(2,3)*Vin ?
(i'll assume there is a reason why.. ie you need to know how the output voltage varies with load, or what the output ripple is..)
or what the input current is.. or SOMETHING!.. )
the answer to that question will direct the non-idealities you need to include.

G'luck!

I am curious about the charge pump dc-dc converter. Is it essentially a combination of diodes and capacitors configured to generate a high DC voltage when driven with an AC input? Is the application primarily RFID chips? Does the converter use any inductors or transformers?

As for the modeling question, Jonathan is quite right in asking about the purpose of the model. If you are interested in dynamics spanning thousands of clock cycles, you may want some sort of state space averaged model. If you are interested in less than 100 clock cycles, or in a model to validate a more abstract model, you probably want a detailed pulse-by-pulse model.

-Jess