One tip you might find useful is to mount all motors or actuators as close to the "base" of the robot arm as you can; essentially, the more mass you put out beyond the base, the more powerful each motor has to be as it gets closer to the base, and the more mass it has to move (and stop!) - leaving less mass for you to lift, as well as making it more difficult to be accurate.
You might do well to take a look at various other small hobbyist robot arms - and I am not talking about the current crop of servo-based arms (though certainly look at them for ideas!). Instead, take a look at this one:http://teachmover.com/
The TeachMover is a classic industrial automation training arm that is still manufactured today (old ones are considered collectible and fairly expensive; new ones are without a doubt very expensive). They originally were designed to be connected via a serial (or sometimes parallel) interface to a home computer or PC - though likely nowadays they use a USB connection of some sort (?). If you do some research on them, you will notice that they use a series of stepper motors located in the base of the arm to control all of the end-effector movements...
Also - take a look at this site:http://www.rhinorobotics.com/xr3_flyer.html
Again - Rhino manufactures a relatively "classic" arm for training purposes - notice the positioning of the motors (in this case, DC motors with integrated encoders, instead of steppers), as well as the chassis design (open elements to reduce the weight of the arm, while enabling high strength). Note, though, that they did position a motor for the gripper at the end of the arm.
Also - take a look at this arm:http://www.senster.com/alex_zivanovic/armdroid/http://theoldrobots.com/clone.htmlhttp://www.anf.nildram.co.uk/beebcontrol/arms/armdroid/
The Armdroid was another one of those "classic" training robot arms - but in this case made in a slightly different fashion from the others to reduce its (comparative) cost significantly, compared to the previous examples. The greatest difference is that the Armdroid used pulleys and cables to route the motion to the joints - instead of the more expensive timing belts and toothed cogs (TeachMover) or chain and sprockets (Rhino). This did, however, cause these arms to fail more often when the cables broke. It's design, though, led to a whole host of "clone" Armdroids being manufactured - mostly in Europe and (at the time) the Eastern Bloc countries; if you happen to find one of these machines being sold, most likely the first thing you'll have to do is "restring" it to get it to work.
The nice thing about the Armdroid, though, is that given the diagrams and information available about it, it should be fairly easy to replicate, if you wanted to do so; the most difficult part would be replicating the differential "wrist" joint, which needs 45 degree bevel gears (an easy way to obtain such gears, though, is to tear down cheap right-angle grinders; many makes have such gears in them).
Finally - on the subject of hobbyist robotic arms - check out this page (if you don't read Spanish - use google translate; it works relatively well for the pages at this site):https://sites.google.com/site/proyectosroboticos/
It might also do you well to examine how regular industrial robots have been constructed throughout the last 60 or so years; there have been many designs and such that you might be able to replicate (the current crop of industrial robots, though, tend to have integrated motors at the joints, which will be difficult to impossible to replicate - such motors are either coupled directly with the joint, or through a harmonic coupling drive to reduce speed and increase torque - all of which makes for a very accurate, though expensive arm). Check out this site for some history of such:http://cyberneticzoo.com/
...as well as googling terms on "industrial robots" and "industrial automation" - and the like...